基于WATD-MTF与改进的残差网络齿轮箱故障诊断研究  

Gearbox fault diagnosis based on WATD-MTF and improved residual network

作　　者：沈景涛 武哲 张强 崔彦平[1] 曹亚超 SHEN Jingtao;WU Zhe;ZHANG Qiang;CUI Yanping;CAO Yachao(College of Mechanical Engineering,Hebei University of Science and Technology,Shijiazhuang 050018,China;National Defense Key Lab of Tank Transmission,China North Vehicle Research Institute,Beijing 100072,China)

机构地区：[1]河北科技大学机械工程学院,石家庄050018 [2]中国北方车辆研究所坦克传动国防重点实验室,北京100072

出　　处：《振动与冲击》2025年第7期247-257,共11页Journal of Vibration and Shock

基　　金：中央引导地方科技发展资金项目(226Z1906G,246Z4502G);河北省教育厅科学研究项目资助(CXY2024038);石家庄市科技局驻冀高校基础研究项目(241791157A)。

摘　　要：针对齿轮箱工作环境复杂多变含噪声大、不同工况下模型泛化性能弱而导致训练准确率低等问题,提出一种小波自适应阈值降噪(wavelet adaptive threshold denoise,WATD)结合马尔可夫转移场(Markov transition field,MTF)与改进的残差网络齿轮箱故障诊断方法。在ResNet18模型的基础上融合了SKNet注意力网络,构成了SK-ResNet18模型,来提高ResNet18模型对重要特征的提取能力。利用WATD算法对一维信号进行去噪,将去噪后的一维信号生成包含时序信息的MTF二维特征图,并输入到改进后的网络中进行特征提取,最终利用网络全连接层实现对故障种类的精确识别。利用东南大学齿轮故障数据集和QPZZ-II试验台采集的齿轮故障数据对该方法进行试验验证,结果表明:该方法能有效识别故障类型,相比其它智能算法,该方法在数据降噪后与不同工况下均表现出较高的优越性和可泛化性能。所提方法可为实际工业的齿轮箱故障诊断任务提供一定的参考价值。Here,aiming at problems of complex and variable working environment with large noise and weak model generalization performance under different operating conditions to cause low training correct rate,a gearbox fault diagnosis method based on wavelet adaptive threshold denoising(WATD)combined with Markov transition field(MTF)and an improved residual network was proposed.SKNet attention network was fused into ResNet18 model to form SK-ResNet18 model,it could improve ResNet18 model’s ability to extract important features.WATD algorithm was used to denoise one-dimensional signals,and the denoised one-dimensional signals could generate MTF two-dimensional feature maps containing time sequence information,and the maps were input into an improved residual network for feature extraction.Finally,the full connected layer of network was used to realize accurate identification of fault types.The proposed method was experimentally verified using Southeast University gear fault dataset and gear fault data collected from QPZZ-II experimental platform.The results showed that the proposed method can effectively identify fault types;compared with other intelligent algorithms,the proposed method exhibits higher superiority and generalization performance under different working conditions and after data denoising;the proposed method can provide certain reference value for practical industrial gearbox fault diagnosis tasks.

关 键 词：故障诊断 SKNet注意力网络 小波自适应阈值降噪(WATD) 马尔可夫转移场(MTF) 残差网络 

分 类 号：TH132.41[机械工程—机械制造及自动化]